Sidekick – a LangGraph‑powered personal co‑worker (Gradio UI)

An AI sidekick that can take multi‑step actions with tools, check its own work against success criteria, and present a friendly Gradio chat UI.

• Runtime & orchestration: [LangGraph] worker ↔ evaluator loop with tool routing
• UI: Gradio chat (+ success criteria box)
• Tools out of the box:
  • Playwright browser toolkit (navigate, scrape, interact)
  • File management (sandboxed read/write)
  • Push notifications (Pushover)
  • Web search (Google Serper)
  • Wikipedia lookup
  • Python REPL
  • Generate images (Pollinations)

---

📷 Sample Interface Screenshot

---

🔐 Environment variables (.env)

Create a .env file in the project root:

OPENAI_API_KEY=sk-...
SERPER_API_KEY=...
PUSHOVER_TOKEN=...
PUSHOVER_USER=...
SERPER_API_KEY=...
LANGSMITH_TRACING="true"
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
LANGSMITH_API_KEY=...
LANGSMITH_PROJECT=...

---

🚀 Installation

Using uv:

uv sync
uv run playwright install

---

▶️ Run

From the project root:

# Run the entry from inside /code
cd code
uv run python app.py
# or: uv run ./app.py  (Windows PowerShell)

---

🧭 How it works

• Worker LLM receives your request + success criteria and can call tools.
• Evaluator LLM checks the last answer against the criteria.
• The graph routes worker → tools → worker → evaluator until:
  • success criteria are met, or
  • the assistant asks a clear question, or
  • a safety turn/recursion limit is reached.

The worker prompt includes a short “tool policy” so it actually uses the tools instead of hallucinating results.

---